A Hybrid 3D CNN and Artificial Ecosystem-based Optimization (AEO) Model for Thyroid Nodule Detection

Generally, doctors frequently require sophisticated diagnostic equipment to identify and do follow-up diagnoses on thyroid nodules. They spend a long time manually extracting features from images and analysing them. Ultrasound images and deep learning-based image detection methods can be applied for thyroid nodule diagnosis. As a strong model for extracting essential features from images, deep learning, especially 3D Convolution Neural Networks (3D CNN), has demonstrated benefits in medical image identification applications. This study employed a hybrid approach of 3D CNN and Artificial Ecosystem-based Optimization (AEO) algorithm for thyroid nodule detection from ultrasound images. A 3D CNN model extracts the features from images. After that, the AEO algorithm is used as a feature picker to identify which traits are most pertinent. The hybrid technique was applied to the dataset that contains 1690 ultrasound images and achieved an accuracy of around 92% compared with 3D CNN, which achieved 88%. In short, this study showed that the hybrid technique between AEO and a 3D CNN model is more accurate for nodule detection than 3D CNN models.